BUG: GH3455 Duplicate indexes with getitem will return items in the correct order

[jreback]

BUG: GH3457 to_csv writing duplicate columns incorrectly

closes #3455, #3457

[ghost]

I'll redo #3458 after you merge this.

[jreback]

ok...any other tests that we should run against this?

[ghost]

I don't think #3457 is fixed by this. I'm happy to just leave it to raise an exception if tried,
engine='python' is just a fallback.

In [11]: df= mkdf(10, 3)
    ...: df.columns = ['a','a','b']
    ...: df.to_csv("/tmp/1.csv",engine='python')
    ...: !cat "/tmp/1.csv"
R0,a,a,b
R_l0_g0,R0C1,R0C1,R0C2
R_l0_g1,R1C1,R1C1,R1C2
R_l0_g2,R2C1,R2C1,R2C2
R_l0_g3,R3C1,R3C1,R3C2
R_l0_g4,R4C1,R4C1,R4C2
R_l0_g5,R5C1,R5C1,R5C2
R_l0_g6,R6C1,R6C1,R6C2
R_l0_g7,R7C1,R7C1,R7C2
R_l0_g8,R8C1,R8C1,R8C2
R_l0_g9,R9C1,R9C1,R9C2

edit: ok, we'll just let #3458 have it raise a warning, and users should transition to the new csv writer code.